Sarkar Abhirup, Santiago Roberto J, Smith Ryan, Kassaee Alireza
Department of Radiation Oncology, Hospital of the University of Pennsylvania, Philadelphia, PA 19104, USA.
Med Dosim. 2005 Spring;30(1):20-4. doi: 10.1016/j.meddos.2004.10.004.
Computed tomgoraphy-magnetic resonance imaging (CT-MRI) registrations are routinely used for target-volume delineation of brain tumors. We clinically use 2 software packages based on manual operation and 1 automated package with 2 different algorithms: chamfer matching using bony structures, and mutual information using intensity patterns. In all registration algorithms, a minimum of 3 pairs of identical anatomical and preferably noncoplanar landmarks is used on each of the 2 image sets. In manual registration, the program registers these points and links the image sets using a 3-dimensional (3D) transformation. In automated registration, the 3 landmarks are used as an initial starting point and further processing is done to complete the registration. Using our registration packages, registration of CT and MRI was performed on 10 patients. We scored the results of each registration set based on the amount of time spent, the accuracy reported by the software, and a final evaluation. We evaluated each software program by measuring the residual error between "matched" points on the right and left globes and the posterior fossa for fused image slices. In general, manual registration showed higher misalignment between corresponding points compared to automated registration using intensity matching. This error had no directional dependence and was, most of the time, larger for a larger structure in both registration techniques. Automated algorithm based on intensity matching also gave the best results in terms of registration accuracy, irrespective of whether or not the initial landmarks were chosen carefully, when compared to that done using bone matching algorithm. Intensity-matching algorithm required the least amount of user-time and provided better accuracy.
计算机断层扫描 - 磁共振成像（CT - MRI）配准常用于脑肿瘤靶区的勾画。我们临床上使用2个基于手动操作的软件包和1个具有2种不同算法的自动软件包：利用骨结构的倒角匹配算法和利用强度模式的互信息算法。在所有配准算法中，在2个图像集的每一个上至少使用3对相同的解剖学标志点，最好是非共面的标志点。在手动配准中，程序对这些点进行配准，并使用三维（3D）变换链接图像集。在自动配准中，将这3个标志点用作初始起点，并进行进一步处理以完成配准。使用我们的配准软件包，对10例患者进行了CT和MRI配准。我们根据花费的时间、软件报告的准确性以及最终评估对每个配准集的结果进行评分。我们通过测量融合图像切片上左右眼球和后颅窝“匹配”点之间的残余误差来评估每个软件程序。一般来说，与使用强度匹配的自动配准相比，手动配准在对应点之间显示出更高的错位。这种误差没有方向依赖性，并且在两种配准技术中，对于较大的结构，大多数时候误差更大。与使用骨匹配算法相比，基于强度匹配的自动算法在配准准确性方面也给出了最佳结果，无论初始标志点是否仔细选择。强度匹配算法所需的用户时间最少，并且提供了更好的准确性。
